The Rundown AI / Articles / AI / OpenAI's spicy new custom AI chip

OpenAI's spicy new custom AI chip

PLUS: Get a deal on World Cup tickets with AI price tracking

Zach Mink

June 25, 2026

Good morning, AI enthusiasts. Back in October, OpenAI and Broadcom said they'd build custom chips together. Nine months later, the first one is running in the lab — and OpenAI's own AI helped build it.

Jalapeño is OpenAI's spicy entry into owning its own compute layer, with "substantially better than current state-of-the-art" performance claims that could give the AI giant a serious efficiency edge.

Reminder — our next live workshop is today at 2 PM EST! Join and learn how to get consulting-grade strategy out of the AI tools you already have, and when it’s worth it to pay up for the top models. RSVP here.

In today’s AI rundown:

OpenAI designs its first custom AI chip
Anthropic, OpenAI join $500M plan to kill colds
Get World Cup ticket deals with AI price tracking
The Fable 5 comeback starts taking shape
4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

OPENAI

🌶️ OpenAI designs its first custom AI chip

Image source: OpenAI

The Rundown: OpenAI just shared new details on Jalapeño, the company’s first custom chip co-built with Broadcom in just nine months to run ChatGPT, Codex, and future agents more efficiently while reducing reliance on Nvidia.

The details:

OAI says Jalapeño went from design to factory-ready in nine months, with its own models helping in design and optimization.
Jalapeño is an ASIC chip that handles inference for running finished models for users, different from the chips used in training.
OpenAI also revealed that Jalapeño has delivered “performance per watt substantially better than current state-of-the-art” in testing.
OpenAI is pushing to power 10 GW of compute with custom chips by 2029, with Nvidia still anchoring model training.

Why it matters: OAI believes the buildout is “the fastest ASIC development cycle ever achieved”, and that’s only going to accelerate as its models become a bigger part of the process. Owning the silicon, models, and products at once lets OAI tune each layer for the other, unlocking cost and speed efficiency gains across the board.

TOGETHER WITH AWS MARKETPLACE

📊 Build strong data foundations for AI

The Rundown: Hear how enterprise leaders at Prudential Insurance, Siemens, GAF, and HF Sinclair build strong data foundations for agentic AI in this virtual panel. Learn how to modernize your data, adopt gen AI, and accelerate solution discovery and deployment with AWS Marketplace.

The panel covers how to:

Build modern data architectures that turn data into an advantage
Power analytics and AI with high-quality, governed data
Adopt gen AI and accelerate solution discovery and deployment
Discover, procure, and deploy AI solutions via AWS Marketplace

INTERCEPT

🦠 Anthropic, OpenAI join $500M plan to kill colds

Image source: Intercept

The Rundown: Stripe, Anthropic, the OpenAI Foundation, and other donors formed Intercept, a $500M nonprofit funding respiratory-virus prevention tools and cleaner indoor-air tech — hoping to make routine sickness like cold and flu “a thing of the past”.

The details:

The group will fund development of shots, sprays, and pills that block dozens of respiratory viruses, plus air-cleaning tech for offices and schools.
Intercept says people lose 15-25 days a year to routine respiratory infections, adding up to roughly $600B in global productivity losses.
The $500M isn't meant to bring products to market, but take early research far enough that pharma firms and investors take over the costly final stretch.

Why it matters: We’ve seen plenty of lofty missions for curing medical issues in the AI-driven science age, but this one hits close to home for everyone. The deep-pocketed AI giants (and their employees) are hoping that funding the awkward middle of development that pharma won’t touch can help get a universal problem over the hump.

AI TRAINING

⚽️ Get World Cup ticket deals with AI price tracking

The Rundown: In this guide, you will learn how to turn ChatGPT Agent Mode into a recurring World Cup ticket scout. Run the search once, tune the prompt, then have ChatGPT check every morning.

Step-by-step:

Open a ChatGPT thread, type /agent, and give a brief detailing your location, the number of tickets needed, and the preferred venues, dates, and budget.
Prompt: “Build a World Cup 2026 ticket scout for [PARTY_SIZE] people from [HOME_CITY]. Include only [DOABLE_VENUES]. Check Ticketmaster, SeatGeek, Vivid Seats, and [OTHER_TICKET_SOURCES]. Return a ranked table with match, date, venue, ticket total, transport, all-in total, link, and next action.”
Run the query once. Verify the result by clicking the source links and making sure the deals look real. Tweak, then turn off Agent Mode.
Now prompt: “Turn this into a recurring task every morning at 8:00 a.m,” and it will run every day, giving you updates on ticket deals.

Pro tip: Try the Codex desktop app for longer scouting runs. Codex is included with paid ChatGPT plans and can work longer while checking more sources.

PRESENTED BY ADAPT

💬 Make your company instantly AI native

The Rundown: The era of solo player AI is over. Adapt is the shared context layer (“company brain”) that enables every employee to work AI native together. Anyone can tag @Adapt in Slack to answer a quick question, schedule a task, prepare for a board meeting, or build the dashboard you’ve been waiting on for weeks.

With Adapt, you can:

Transform your Slack into a shared AI workspace
Connect to all of the tools you already use
Easily share workflows, skills, and knowledge across teams
Experience AI built for business users and powerful enough for engineers

Get started with $100 free credits — and ask about our pilot program, exclusively for The Rundown subscribers.

ANTHROPIC

👀 The Fable 5 comeback starts taking shape

Image source: Images 2.0 / The Rundown

The Rundown: Anthropic's Fable and Mythos remain offline in compliance with the U.S. order, but there's smoke surrounding their path back — with Claude Code strings, White House talks, and a lawsuit hinting that the freeze may be starting to thaw.

The details:

A new Claude Code update log includes references to Fable usage and the removal of ‘purchased separately’, pointing to potential usage changes.
WIRED reported that the Trump admin is “happier” in talks with Anthropic, now dealing with co-founder Tom Brown instead of Dario Amodei in meetings.
Legal-tech firm Legion filed the first lawsuit against the order, calling it "unlawful" and arguing "the government’s real motive was unlawful retaliation."
Reps. Sam Liccardo, Jay Obernolte, and Ted Lieu gave Commerce until June 26 to lay out how and when the public could get the model back.

Why it matters: Where there is smoke, there is fire, and hopefully that means Fable’s return to the frontier is coming soon after nearly a month on ice. It sounds like removing Anthropic’s outspoken CEO from talks is working, given this quote from WIRED: “Tom Brown is not being a weirdo like Dario and can actually engage”.

QUICK HITS

🛠️ Trending AI Tools

💬 Claude Tag - Tag @Claude as a shared Slack teammate to delegate tasks
🎬 OpenArt Director - Vibe-directing tool that builds 5-minute films from chat
🎨 Krea 2 - Krea’s open weights in-house image model
📃 OCR 4 - Mistral’s OCR model with layout-aware document understanding

📰 Everything else in AI today

Cube on Databricks Webinar, June 30 – meet Cube, the BI and Agentic Analytics platform – powering dashboards, embedded analytics, and AI agents on Databricks. Learn more and register for free.*

Google AI researchers Jonas Adler, Alexander Prtizel, and Arthur Conmy are leaving the company and joining Anthropic, the latest in a wave of departures.

Former Anthropic, OpenAI, Google, and xAI employees launched Mirendil, a new lab building AI that runs its own R&D for self-improving science research.

Google DeepMind added new computer-use functionality directly into Gemini 3.5 Flash, allowing the model to power agents that navigate browsers, apps, and desktops.

xAI’s Grok is heavily driven by NSFW generations, The Information reported, with xAI employees saying that “well over half” of its traffic comes from explicit content.

*Sponsored Listing

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Steve C. in Troy, NY:

"A young member of my family has selective mutism. At home, she is completely comfortable, but in social situations… speaking can be really hard for her. I wanted to see if I could build something that would help her practice in a low-pressure way.

Using Claude, I built a voice-based practice app from scratch. It lets her choose who she wants to practice with, such as a kid her age, a teacher, or a coach, and pick a real-world situation like a cafeteria.

I added background audio so the setting feels more realistic. The character asks her age-appropriate questions out loud, she responds using her voice, and Claude gives her warm, honest feedback. Parents can adjust the feedback style and the number of questions in each session. She is now using it, and that is what makes this one feel different for me. This was not about building something complex. It was about using AI to solve a real problem for someone I care about.”

How do you use AI? Tell us here.